From fc79b980a0d4f0b219585c55402690b92c3dddf8 Mon Sep 17 00:00:00 2001 From: Asif P Date: Thu, 22 Jan 2026 01:17:56 -0700 Subject: [PATCH] [DOC] Add warning about over-sampling limitations --- doc/over_sampling.rst | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) diff --git a/doc/over_sampling.rst b/doc/over_sampling.rst index 3bc975b89..de6b1ffb8 100644 --- a/doc/over_sampling.rst +++ b/doc/over_sampling.rst @@ -1,4 +1,4 @@ -.. _over-sampling: +oo.. _over-sampling: ============= Over-sampling @@ -12,6 +12,17 @@ A practical guide You can refer to :ref:`sphx_glr_auto_examples_over-sampling_plot_comparison_over_sampling.py`. +.. warning:: + + Over-sampling methods (including random over-sampling, SMOTE, and ADASYN) do + not guarantee improved predictive performance. In particular, they may not + improve ranking metrics such as ROC-AUC and can degrade probability + calibration. We recommend validating the impact using the metric(s) that + matter for your use case, and considering alternatives such as using + ``class_weight`` or ``sample_weight`` in the estimator, tuning the decision + threshold, or collecting more minority class data when possible. + + .. _random_over_sampler: Naive random over-sampling